NemoClaw Knowledge Wiki

PDP Gary AI Integration, Retro Computing Projects, and AI-Driven Code Optimization

6 min read

🗂️ Tools, Platforms & Infrastructure · View mindmap

PDP Gary AI Integration, Retro Computing Projects, and AI-Driven Code Optimization

Clip title: Meet PDP Gary: AI Meets Old-School Computing | Shop Talk #98 Author / channel: Dave’s Attic URL: https://www.youtube.com/watch?v=PsmdflXfY2U

Summary

This video, “Shop Talk,” features Dave answering questions and discussing recent projects and insights related to computing, AI, and his custom task manager, TMOG. The main topics span from the development and capabilities of his AI assistant and porting efforts on retro systems, to deeper discussions on AI limitations and the practicalities of system monitoring tools.

Dave elaborates on “Gary,” his AI companion integrated into a VT220 terminal setup. Gary’s personality is shaped by an extensive text file, outlining specific preferences and attitudes, such as a fondness for the PDP-11/70 and a bias towards 1980s solutions. While Gary leverages Quinn 3.8 flash on the backend, the user interface and interaction are managed by a physical PDP-11, which dispatches requests to GPUs over Ethernet and processes the AI’s responses. Furthermore, Dave discusses his ongoing work on a PDP-11/70, specifically building a BSD kernel for a new machine without a floating-point processor and overcoming challenges with drive controller memory addresses. He also successfully ported KSH (KornShell) to 2.11BSD, improving upon the older TCSH and finding Bash too resource-intensive for the retro system.

A significant portion of the discussion revolves around using AI for coding and troubleshooting. Dave describes how he challenged AI to write the “top” command for BSD Unix. After initial feedback to remove debug symbols and further instructions to optimize for “lean and fast” execution, the AI-generated code surpassed the original hand-coded version in efficiency and startup time. He also recounts using ChatGPT to diagnose a subtle error in a 7-digit octal memory address for a peripheral controller on his PDP-11/70, which saved considerable physical troubleshooting effort. Beyond coding, the conversation delves into AI’s broader implications, emphasizing the necessity of strict “guardrails” to prevent autonomous agents from engaging in undesirable or harmful actions. He stresses the need for clear definitions of “alignment” in AI development to ensure ethical behavior and highlights that while AI can generate complex plans (like house blueprints), human oversight and rigorous testing remain critical for validating accuracy and safety. He also clarifies that consistent AI responses to identical prompts can be achieved by controlling the model’s “temperature” (randomness) and using a fixed seed, though public models often introduce variability.

Finally, Dave details his custom task manager, TMOG (Task Manager OG), contrasting it with traditional tools. TMOG aims to unify various system performance metrics—such as CPU, memory, thermal, and network data—into a single, consolidated view, making it easier to diagnose system slowdowns. The development of TMOG involved a detailed specification and AI assistance for generating boilerplate code, with human developers focusing on refining its functionality and ensuring minimal system overhead (consuming only 1-3% CPU). TMOG includes a “Flight Recorder” feature to log system data over extended periods for post-mortem analysis of issues like crashes. It also offers a thermal process page to identify applications consuming significant power. TMOG supports cross-platform use (Windows, macOS, Linux) and can replace the default task manager in Windows. Dave continues to enhance TMOG based on user feedback and personal interest, viewing it as an ongoing pursuit rather than a finished product.

Video Description & Links

Description

PDP Gary has a personality—but how did Dave define it? This week on Shop Talk, Dave and Glen start with Gary, then explore Dave’s recent work using AI on BSD, the PDP kernel, and its peripherals.

From there, we turn to viewer questions about AI control, trust, and why the same question can produce different answers.

We also dig into Task Manager OG: what it adds beyond Windows Task Manager, tracking down resource spikes, memory usage versus memory pressure, and using Flight Recorder to investigate crashes. Plus, cross-platform C++, AI-assisted development, and questions about licensing and TMOG’s future.

What would you ask PDP Gary? Leave your question in the comments.

00:00 - Hey I’m Dave… 00:33 - Defining PDP Gary’s Personality 02:10 - AI, BSD and PDP Kernel Work 08:10 - Stopping an Unstoppable AI Agent 09:23 - Proving Claims About AI Actions 11:17 - Would You Trust AI to Design Your House? 13:35 - Same AI Question, Different Answers 16:04 - TMOG vs. Task Manager 16:55 - Why Did It Take 40 Years? 17:39 - TMOG vs. Process Explorer 19:13 - Does TMOG Replace Task Manager? 19:57 - Which Process Caused the Spike? 21:35 - Memory Usage vs. Memory Pressure 22:28 - Does Monitoring Add Bloat? 23:08 - CPU Work vs. Overhead 24:16 - Tracking a Maxed-Out CPU Core 24:54 - Audio Glitches, Frame Drops and Core Scheduling 27:33 - What Caused the Throttling and Fan Activity? 28:26 - Diagnosing Crashes with Flight Recorder 29:40 - Threshold-Triggered Flight Recording 31:01 - One C++ Core Across Three Operating Systems 31:31 - How Much of TMOG Was AI-Coded? 33:20 - TMOG’s Windows UI and Phosphor Look 34:40 - TMOG Source Code and Microsoft Ownership 35:06 - Why Does TMOG Connect to the Internet? 35:47 - Licence Checks and Lifetime Updates 36:17 - Code Signing and Task Manager Shortcuts 38:42 - Portable Flight Recorder for Remote Diagnostics 39:34 - TMOG Business Licensing 40:31 - Would Dave Sell TMOG to Microsoft? 42:08 - TMOG for Servers and Data Centres 42:33 - Remote Monitoring on Headless Linux 43:03 - More System Tools After TMOG? 44:36 - Getting Lost in Winamp Skins 45:37 - Dave’s Norm Macdonald Resemblance 46:25 - Why Does Bobcat Goldthwait Owe Dave Money? 48:52 - Shenanigans and outtakes

Graph View